Intelligent Embedded Systems
This course focuses on the design, implementation, and deployment of AI/ML driven edge computing systems in real-time, resource-constrained environments. Students will explore modern artificial intelligence (AI) and machine learning (ML) techniques, implement algorithms on dual-core STM32H7 microcontrollers, and gain hands-on experience deploying PyTorch-trained models (exported to ONNX) using STM32Cube.AI. The course integrates concepts from AI/ML, embedded systems, and energy-efficient programming to build end-to-end edge applications.
While AI/ML models are commonly deployed in high-level programming environments such as Python or on single-board computers running Linux and Python interpreter, real-time embedded systems often require a different approach. This course emphasizes the deployment of AI/ML models on resourceconstrained microcontrollers using low-level programming in C/C++. Such an approach ensures realtime performance, deterministic execution, and energy efficiency—critical factors for applications in robotics, automotive systems, and industrial automation. Students will learn how to optimize models for embedded execution, efficiently manage memory and compute resources, and implement inference pipelines without relying on an operating system. By writing software in C/C++, students will ensure real-time performance, deterministic execution, and energy efficiency.
The STM32 Model Zoo will be a key resource in our course, enabling students to explore AI applications on embedded devices and integrate machine learning models into real-world chip design projects. It will provide hands-on experience with deploying and customizing machine learning models on low-power, resource-constrained microcontrollers, an essential skill in modern embedded system development. The STM32 Model Zoo is a collection of pre-trained machine learning models optimized for deployment on STM32 microcontrollers. Developed by STMicroelectronics, it provides a valuable resource for integrating edge AI capabilities into embedded systems. This tool offers models tailored for various applications, such as image classification, object detection, and audio recognition, and is particularly suitable for use with STM32 microcontrollers, including those with the Neural-ART Accelerator (NPU).
We plan to utilize the STM32N6 and STM32H7 series in the course to demonstrate the practical application of AI models and machine learning on embedded microcontrollers, providing students with hands-on experience in deploying efficient AI solutions on low-power devices. The STM32N6 series is a family of microcontrollers from STMicroelectronics, designed specifically for AI and machine learning applications at the edge. These microcontrollers are equipped with the Neural-ART Accelerator, a hardware-based accelerator optimized for running machine learning algorithms efficiently. The STM32H7 dual-core series is also an excellent, low-cost platform for deploying AI and machine learning algorithms in embedded systems. Its unique dual-core architecture, combining a high-performance Arm Cortex-M7 core and a lower-power Arm Cortex-M4 core, enables efficient handling of both complex AI tasks and resource-constrained operations. This flexibility makes the STM32H7 series ideal for edge AI applications,
where real-time decision-making and efficient resource management are crucial.
